workflow_run
Run a sequence of browser automation steps (navigate, click, fill, etc.) with error handling and resume from a specific step if failure occurs.
Instructions
⭐ Execute a list of tool steps sequentially. Resumable — pass start_at=N to skip the first N steps.
Each step: {"tool": "<name>", "args": {...}, "label": "optional"}
Args:
steps: list of step dicts
start_at: index to begin from (for resume after a fix)
stop_on_error: abort on first failure (default True). If False,
continue and collect all results.
Returns JSON:
{
"completed": [
{"index": 0, "tool": "navigate", "ok": true, "result": "..."},
...
],
"failed_at": 3, # index of failure (omitted on success)
"failure_context": {...}, # last step's input + error (for LLM debug)
"resume_with": "workflow_run(steps=..., start_at=4)" # hint
}
Allowed tools (curated for sequencing): navigate, reload, go_back/forward,
click, click_text, click_role, fill, type_text, press_key, select_option,
check, uncheck, wait_for*, screenshot, scroll, scroll_to, smart_fill,
vision_locate, assert_*, storage_*, cookie_import, storage_state_load,
evaluate, mouse_click_xy.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| steps | Yes | ||
| start_at | No | ||
| stop_on_error | No |
Output Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| result | Yes |